why doesn't RAID performance scale? - Storage

This is a discussion on why doesn't RAID performance scale? - Storage ; Maybe some of the folks here can answer this... Why doesn't RAID performance scale? If a 7200RPM disk can write 20MB/s writes, why can't a stripe of 5 give 100MB/s? If I use a controller with HW parity, surely it ...

+ Reply to Thread
Results 1 to 3 of 3

Thread: why doesn't RAID performance scale?

  1. why doesn't RAID performance scale?

    Maybe some of the folks here can answer this...

    Why doesn't RAID performance scale?

    If a 7200RPM disk can write 20MB/s writes, why can't a stripe of 5
    give 100MB/s?

    If I use a controller with HW parity, surely it should be able to keep
    up with a 6 disk RAID 5 volume and perform similar to a non-parity
    stripe.

    This is not just a single vendor...

    I get about 60MB/s on a Sun 3511 array (big bucks)
    I get around 45MB/s on a HP P400 (w/battery cache)
    I've noticed similar problems on a Dell controller (don't remember the
    model#)

    In all cases, the controller write cache is enabled of course.

    Shouldn't RAID scale?

    Robert

  2. Re: why doesn't RAID performance scale?

    Previously rmeden@yahoo.com wrote:
    > Maybe some of the folks here can answer this...


    > Why doesn't RAID performance scale?


    It does.

    > If a 7200RPM disk can write 20MB/s writes, why can't a stripe of 5
    > give 100MB/s?


    If it is a RAID1 stripe and the rest of the system can handle 100MB/s
    wor mixed device access, you will get that performance.

    > If I use a controller with HW parity, surely it should be able to keep
    > up with a 6 disk RAID 5 volume and perform similar to a non-parity
    > stripe.


    No. Some hardware controllers are surprisingly slow.

    > This is not just a single vendor...


    > I get about 60MB/s on a Sun 3511 array (big bucks)
    > I get around 45MB/s on a HP P400 (w/battery cache)
    > I've noticed similar problems on a Dell controller (don't remember the
    > model#)


    First, not only the controller is a bottleneck. Second
    is this reading or writing? And what is the performance of the
    individual disks, when connected to this contoller?

    > In all cases, the controller write cache is enabled of course.


    > Shouldn't RAID scale?


    As I said, it does. But hardware has kimuts. It does not scale
    past these limits.

    Here is a speed sample from a 8-way Arcea RAID6:

    Read speed 340MB/s. That is 56.6MB/s for the individual disks.
    I believe the disks are in the 60MB/s class. Looks like good
    scaling to me. Maybe your hardware is PCI or is just over-priced?

    Arno

  3. Re: why doesn't RAID performance scale?

    rmeden@yahoo.com wrote in news:e3133f6c-32d3-4344-a770-9f1fe3b77528@b1g2000hsg.googlegroups.com
    > Maybe some of the folks here can answer this...


    > Why doesn't RAID performance scale?


    It doesn't?

    >
    > If a 7200RPM disk can write 20MB/s writes, why can't a stripe of 5
    > give 100MB/s?


    It can't?

    >
    > If I use a controller with HW parity,


    > surely it should be able to keep up


    Whatever 'it' is supposed to mean.

    > with a 6 disk RAID 5 volume


    And why is that?

    > and perform similar to a non-parity stripe.


    You mean, to RAID0? Didn't you just say that 'RAID' doesn't scale?

    RAID5 writes of incomplete strip(e)s (or whatever a full complement of
    blocks is called here) also require a read in order to calculate the parity.
    This should not showup though in static benchmarks.

    >
    > This is not just a single vendor...
    >
    > I get about 60MB/s on a Sun 3511 array (big bucks)
    > I get around 45MB/s on a HP P400 (w/battery cache)
    > I've noticed similar problems on a Dell controller (don't remember the
    > model#)


    > In all cases, the controller write cache is enabled


    > of course.


    Of course.
    And now you are surprized that you are measuring the cache performance?

    >
    > Shouldn't RAID scale?


    Doesn't it?

    >
    > Robert



+ Reply to Thread